1. The Field of the Invention
The present invention relates to systems and methods for transferring data to and from memory in a computer system. More particularly, the present invention relates to systems and methods for servicing the data and memory requirements of system devices by arbitrating the data requests of those devices.
2. The Prior State of the Art
An important operational aspect of a computer or of a computer system is the need to transfer data to and from the memory of the computer. However, if the computer's processor is used to perform the task of transferring data to and from the computer's memory, then the processor is unable to perform other functions. When a computer is supporting high speed devices that have significant memory needs, the processor bears a heavy load if the processor is required to copy data word by word to and from the computer's memory system for those devices. As a result, using the processor to transfer data in this manner can consume precious processing time.
A solution to this problem is Direct Memory Access (DMA). A DMA controller essentially relieves the processor of having to transfer data to and from memory by permitting a device to transfer data to or from the computer's memory without the use of the computer's processor. A significant advantage of DMA is that large amounts of data may be transferred before generating an interrupt to the computer to signal that the task is completed. Because the DMA controller is transferring data, the processor is therefore free to perform other tasks.
As computer systems become more sophisticated, however, it is becoming increasingly evident that there is a fundamental problem between the devices that take advantage of DMA and the memory systems of those computers. More specifically, the problem faced by current DMA modules is the ability to adequately service the growing number of high speed devices as well as their varying data requirements.
High performance memory systems preferably provide high bandwidth and prefer large data requests. This is in direct contrast to many devices, which may request small amounts of data, have low bandwidth, and require small latencies. This results in system inefficiencies as traditional devices individually communicate with the memory system in an effort to bridge this gap. It is possible that many different devices may be simultaneously making small data requests to a memory system that prefers to handle large memory requests. As a result, the performance of the memory system is decreased.
This situation makes it difficult for low bandwidth devices, which may have high priority, to effectively interact with high bandwidth devices that may have lower priority. For example, an audio device may support several different channels that receive data from memory. The audio device typically makes a data request to memory for data every few microseconds for those channels. Because devices such as audio devices recognize that they may experience significant latency from the memory system before their request is serviced, the audio device, may implement an excessively large buffer to account for that latency.
This is not an optimum solution for several reasons. For instance, many devices maintain a large buffer because they do not have a guarantee that their data requests will be serviced within a particular time period. Other devices maintain an excessively large buffer because it is crucial that the data be delivered in a timely manner even though the devices may have low bandwidth requirements. For example, if an audio device does not receive its data in a timely manner, the result is instantly noticed by a user. Additionally, each device must implement DMA control logic, which can be quite complex for some devices. In other words, the DMA control logic is effectively repeated for each device.
Current devices often interact with DMA systems independently of the other system devices and each device in the system is able to make a data request to the DMA at any time. As a result, it is difficult to determine which devices need to be serviced first. The arbitration performed by systems employing isochronous arbitration often defines fixed windows in which all devices that may require servicing are given a portion. These fixed windows are large from the perspective of high bandwidth devices and small from the perspective of low bandwidth devices. Thus, high bandwidth devices are required to buffer more data than they really need and low bandwidth devices often do not need to use their allocated portion of the window. This results in inefficiencies because all of the available bandwidth may not be used and additional memory is required for the buffers of high bandwidth devices. In essence, current systems do not adequately allow high priority devices to efficiently coexist with high bandwidth devices.